NIST’s image standards could become win/lose factor for govt biometrics contractors
The U.S. government is developing a vendor-neutral standard for assessing how useful an image might be for biometric identification. The face image quality standard could be set by the start of 2024.
According to reporting by Bloomberg Law, the goal is to show facial images to Transportation Security Administration and Customs and Border Protection agents that are more accurate and useful and less biased.
The Department Homeland Security (which oversees the TSA and CBP) and the National Institute for Standards and Technology reportedly are working together on the tool.
Machine vision vendors right now find it difficult to judge how their image capture technology for use with biometric systems will satisfy the government’s needs.
As basic as it sounds, demonstrating proper lighting through the standard could, according to Bloomberg, eliminate at least some of the image bias against darker skin tones.
Idemia North America CEO Donnie Scott is quoted in the article saying the government has used NIST well as a technology evaluator. Scott is referring in part to the well-known and -respected Face Recognition Vendor Test, an exhaustive and ongoing program of testing biometric algorithms for strengths and faults – among them being demographic bias.
Scott said NIST’s work needs to be carried forward, though. Adherence to NIST standards should be a box to tick for contractors bidding on government work.
And while not everyone thinks this idea is comprehensive enough to celebrate, NIST is planning its first full AI Risk Management Framework, according to Bloomberg.
Its goal is to show coders how best to address five critical factors in deploying facial recognition or other AI techniques: bias, security, explainability, reliability and accuracy.
accuracy | biometric identification | biometric-bias | biometrics | explainability | Face Recognition Vendor Test (FRVT) | facial recognition | NIST | research and development | standards | U.S. Government